Application of The K-Means Clustering Algorithm in Medical Claims Fraud / Abuse Detection

International Journal of Application or Innovation in Engineering& Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org

Volume 3, Issue 7, July 2014 ISSN 2319 - 4847

Volume 3, Issue 7, July 2014 Page 142

ABSTRACT
This paper is about a system which applies a modified K-Means algorithm[12] to flag out suspicious claims for further
scrutiny has been developed. The Java programming Language and mySQL database tools were used. The K-Means algorithm is
well known for its efficiency in clustering large data sets. However, a major limitation of this algorithm is that it works only with
numeric values, thus the method cannot be used to cluster real-world data containing categorical values. To counter this, data
sets were converted to numeric data whereby ailments were listed and matched with patients. The presence of the ailment was
represented by a one (1) and the absence was represented by a zero (0). To get the data, a total of 15 insurance companies in
Kenya out of 31 were randomly selected and a pre-tested questionnaire was used to collect data. 15 insurance companies out of
31 is close to 50%, which is a very good representative of the entire population. 67 % of the respondents indicated that the people
involved in the processing of claims were billing for services that were not rendered. The results also showed that all the
companies had internal control mechanisms to address the problem and 47% of the respondents said the internal controls were
not efficient. 87% of the respondents indicated that the common member fraud cases involved mmembership substitution
including card abuse.
Keywords: billing , K- Means, 0s and 1s, clustering, Euclidean distance
1. INTRODUCTION
To develop a medical fraud detection system applying the K-Means [12], the data collected were converted to 0s and 1s
and the Euclidean distances [14],[16] were computed and these distances were used to cluster given data sets. The
average claim amount for a given cluster was computed and claims that very high figures far away from the computed
average claim within that cluster were flagged for further scrutiny or rejected altogether. Clustering [1],[10] is a
fundamental operation in data mining[8]. It is useful in a number of tasks, such as classification, aggregation and
segmentation or dissection. For example, by partitioning objects into clusters, interesting object groups may be
discovered, such as the groups of motor insurance policy holders with a high average claim cost [3], or the groups of
clients in a banking database having a heavy investment in real estate. Clustering [1],[10],[12]. is a popular approach to
implementing the partitioning operation. Clustering methods partition a set of objects into clusters such that objects in
the same cluster are more similar to each other than objects in different clusters according to some defined criteria.
Statistical clustering methods [11] partition objects according to some (dis)similarity measures, whereas conceptual
clustering methods cluster objects according to the concepts objects carry [9] data sets. The data sets to be mined often
contain millions of objects described by tens, hundreds or even thousands of various types of attributes or variables
(interval, ratio, binary, ordinal, nominal, etc.). This requires the data mining operations and algorithms to be scalable
and capable of dealing with different types of attributes. This paper focuses on the K-Means [12] clustering method with a
minor modification to achieve more efficiency. This technique is efficient for processing large data sets. Therefore, it is
best suited for data mining. However, the K-Means algorithm[12] only works on numeric data, i.e., the variables are
measured on a ratio scale , because it minimises a cost function by changing the means of clusters. This prohibits it from
being used in applications where categorical data are involved. Unlike statistical clustering methods, the K-Means
algorithm [12] is based on a search for objects which carry the same or similar concepts. Therefore, its efficiency relies
on good search strategies. Setting the number of clusters before hand is another major drawback when using the K-
Means algorithm since it involves guess work as there is no formula that can be used to calculate the exact number of
clusters required for a given data. The K-Means algorithm is well known for its efficiency in clustering large data sets.
The algorithms limitation of working on numeric data only was addressed by converting the data sets to numeric data
whereby ailments were listed and matched with patients. The presence of the ailment was represented by a one (1) and
the absence was represented by a zero (0) and then these numeric data are fitted in the Euclidean formula for distance
measure which is used in clustering the records.

Application of The K-Means Clustering
Algorithm In Medical Claims Fraud / Abuse
Detection
1
Leonard Wafula Wakoli, Abkul Orto
2
and Stephen Mageto
3
1
Department of Computer Science, Kenya Methodist University - Kenya

2
Department of Information Technology, Meru University of Science and Technology - Kenya

3
Department of Information Technology, Meru University of Science and Technology - Kenya



2. LITERATURE REVIEW
The K-Means Clustering Technique
The K-Means clustering technique is a non-hierarchical approach used to form good clusters by specifying a desired
number of clusters, say, k, then assigning each case (object) to one of k clusters so as to minimize a measure of dispersion
within the clusters [2] A very common measure is the sum of distances or sum of squared Euclidean distances from the
mean of each cluster [5]. The problem can be set up as an integer programming problem but because solving integer
programs with a large number of variables is time consuming, clusters are often computed using a fast, heuristic method
that generally produces good (but not necessarily optimal) solutions. The K-Means algorithm is one such method. The K-
Means method is a very popular algorithm for clustering high-dimensional data. Initiated with k arbitrary cluster centres,
it assigns every data point to its nearest center, and then readjusts the centers, reassigns the data points, until it stabilizes.
The K-Means method terminates in a local optimum, which might be far worse than the global optimum. However, in
practice it works very well. It is particularly popular because of its simplicity and its speed: In practice, the number of
iterations is much less than the number of samples[6].The K-Means clustering algorithm divides the data set into a
predetermined number, k, of clusters. These clusters are centred at random points in the record space[4][7]. Records are
assigned to the clusters through an iterative process that moves the cluster means (also called cluster centroids) around
until each one is actually at the centre of some cluster of records. (See Figures 2.1, 2.2 and 2.3

In the first step, k data points are selected to be the seeds more or less arbitrarily. Each of these seeds is an embryonic
cluster with only one element. In the example shown in figure 1, k is 3.

Figure 2.2: Initial clusters and inter-cluster boundaries (from Berry & Linoff 2000).

In the second step, each record is assigned to the cluster whose centroid is nearest to that record. This forms the three
clusters shown in Figure 2.2 with the new inter-cluster boundaries. Note the boxed record which was assigned to cluster 2
(Seed 2) initially now becomes part of cluster 1

Seed 1
Seed 2
Seed 3
Figure 2.1 Initial cluster seeds (from Berry & Linoff 2000).
Seed 1
Seed 2
Seed 3
Seed 1
Seed 2
Seed 3
Figure 2.3: New clusters, their centroids
marked by crosses and inter-cluster


2.3 Related Efforts
In [17], an on-line discounting learning algorithm to indicate whether a case has a high possibility of being a statistical
outlier in data mining applications such as fraud detection is used for identifying meaningful rare cases in health
insurance pathology data from Australias Health Insurance Commission (HIC). The performance of a k -Nearest
Neighbor (kNN) algorithm with the distance metric being optimized using a genetic algorithm was applied in a real
world fraud detection problems faced by the HIC. In Chile, a single neural network to detect fraudulent medical claims
was implemented in another healthcare insurance company. This scheme utilizes all the data available in arriving
medical claim for constructing a unique vector which is evaluated by the single neural network.
3. METHODOLOGY
The methodology describes the research design, subjects, target population, sampling procedures and sample size
informed consent, Research instruments, validity of the Research instrument, reliability of the instrument, Data
collection, Survey results and data analysis the statistical techniques that were used to analyse the data collected as well as
software development tools. Potential problems, limitations and ethical considerations are presented.
Research Design
Research design can be thought of as the structure of research -- it is the "glue" that holds all of the elements in a research
project together. It is the investigators action plan for answering the research questions and realizing the objectives [15].
Survey designs are often known as correlational designs to denote the tendency to reveal relationships between variables
and to draw attention to their limited capacity in connection with elucidation of casual processes. The design was very
appropriate because it was able to elicit a diverse range of information about the area of study.
Subjects
The unit of analysis is the study fraud cases in the health care sector in Kenya. Focus is based on the assumption that
medical claims can be clustered using the K-Means technique and detecting fraud using the Euclidean distance measure.
Target population
The thirty six (36) active insurance companies in Kenya
Sampling procedures and Sample size
Subjecting the whole target population to investigation is usually an impossible taskas a result of prohibitive costs and the
time involved. This then calls for a sample, which is a subset of the target population through which the requisite
information can be obtained at reasonable costs [15]. Samples should be as representative as possible, because too-small-
a-sample is likely to yield under-estimated information that may not reflect the actual population characteristics or
perceptions . In situations where a population is too small to be sampled, it is logical to sample all the elements [15].
This study applied random sampling procedure to obtain samples of the insurance companies as the ultimate units of
study. Out of the target population of 36 insurance companies, 15 were randomly selected for study, being close to half
the total population of Insurance companies in Kenya As proposed in Chapter one, this study was designed to examine
medical insurance claim fraud/abuse cases at various service providers in Kenya and the subsequent designing and
development of a system to address the problem.
Research Instrument
The study had only one set of the research instrument which focused the establishment of the extent of medical fraud in
Kenya, current mechanisms of detecting the medical claim fraud and the root causes of this type of fraud. The instrument
had 15 questions
Validity of the Research instrument
Validity of research instruments is a key element of an accomplished study. It denotes the extent to which an instrument
captures it purports to measure[15]. The acceptable level of validity largely depends on logical reasoning, experience and
professionalism of the investigator [16]. Pilot testing is a crucial step in the research process because it helps in refining
instruments so that they capture the intended data. Pilot testing reveals what works and what do not, for example, vague
questions and unclear instructions[13].
Reliability of Instruments/tools of the survey
The only instrument/tool that was used to collect data was the Questionnaire which was field-tested after it was designed.
The questionnaire included a mix of check -the-box items, fill-in blanks and closed ended questions.
Data Collection
This started with the designing of the research instrument and recruitment of one research assistant to assist with data
collection activities. This followed consensus building involving the investigator and the research assistant. The purpose
was to discuss the items contained in the instrument for familiarity, logical requirements and acquisition of the research
permit. The data from each of the questions were summarized and presented in summary tables then subjected to
statistical analysis techniques.



4. FINDINGS AND DISCUSSION
4.1 Findings
Common types of fraud that make claims to be rejected:
A. Billing for services that were not rendered.
B. Billing for more expensive services than were actually provided
C. Performing medically unnecessary services so as to generating insurance payments
D. Falsifying a patients diagnosis to justify tests, surgeries or other procedures that are not medically necessary
E. Double billing
F. Other (specify)
Table 4.1: Common reasons for claim rejection
Reason for Claim
Rejection
No. of
respondents
% age of
respondents
A 10 67
B 8 53
C 3 20
D 9 60
E 2 13
F 1 06

Bar chart for Reasons for Claim Rejection
67
53
20
60
13
6
0
10
20
30
40
50
60
70
80
A B C D E F
Reason for Claim Rejection
%

a
g
e

o
f

r
e
s
p
o
n
d
e
n
t
s
% age of respondents

Figure 4.1: Common Reasons for Claims Rejection

Figure 4.1 shows that billing for services that were not rendered and falsifying a patients diagnosis to justify tests,
surgeries or other procedures that are not medically necessary are the most common fraudulent activities that cause the
rejection of medical claims.
How companies determine fraudulent claims:

Control No. of Respondents % age of Respondents
Internal RiskControl 15 100
By chance 13 87
Other 5 33

Bar chart for Fradulent Claim Determination
100
87
33
0
20
40
60
80
100
120
Internal Risk Control Bychance Other
Control
%

a
g
e

o
f

R
e
s
p
o
n
d
e
n
t
s
% age of Respondents

Figure 4.2: Fraudulent Claim determination


The results show that most of the insurance companies have put in place internal controls to try to address the issue of
medical claim fraud
4: Efficiency of internal controls in fraud detection:
Table 4.3: Efficiency of Internal control mechanisms

Pie Chart for Efficiency of internal controls in fraud detecti on
47%
53%
Efficient
Not efficient

Figure 4.3: Efficiency of Internal control mechanisms
Method(s) used for uncovering fraud in organisation
A. Internal controls
B. Internal Audit
C. Notification of employee
D. Accident
E. Anonymous tip
F. Notification by customer
G. Notification by a Regulatory or Law enforcement agency
H. Notification by Vendor
I. External Audit
Table 4.4: Methods for uncovering fraud
Method for
uncovering fraud
No. of
Respondents
% age of Total No of
respondents
A 15 100
B 15 100
C 9 60
D 7 47
E 8 53
F 8 53
G 4 27
H 3 20
I 5 33
Bar chart for methods for uncovering fraud
100 100
60
47
53 53
27
20
33
0
20
40
60
80
100
120
A B C D E F G H I
Method for uncovering fraud
%

a
g
e

o
f

T
o
t
a
l

N
o

o
f

r
e
s
p
o
n
d
e
n
t
s
% age of Total No of respondents

Figure 4.4: Methods for uncovering fraud

Type of Response No. of Respondents % age of Respondents
Efficient 7 47
Not efficient 8 53
Total 15 100


Internal controls and internal audits were found to be the most commonly used methods for uncovering fraud.
4.2 Summary of the Results Analysis
The results clearly show that insurance companies in Kenya are finding it difficult to address the problem of Medical
Claim Fraud. However, most of the companies are trying to put up a spirited fight to combat the said fraud using various
approaches to combat the Insurance Claims Fraud.
4.3 Clustering using the Euclidean distances concept
a) N-dimensional distance
The Euclidean distance between points P =(p
1
, p
2
, ,p
n
) and Q =(q
1
,q
2
, q
n
), in Euclidean n-space is defined as:

((p
1 -
q
1
)
2
+(p
2
q
2
)
2
++(p
n
q
n
)
2
) = (p
i
q
i
)
2
(1)

Using this Euclidean algorithm, medical claim records are clustered using the distance measure technique.
Consider two records: Record 1 and Record 2 as shown in (b) below
b) Euclidean and Manhattan Distance between two records
Table 4.5 : Euclidean and Manhattan Distance between two records

Variable

Record 1
sore throat
swollen neck,
and fractured shoulders
Record 2
Absolute Difference
Square
ruptured toe
and broken elbow

Absolute Difference Square

Shoulder 1 0 1 1
Knee 0 0 0 0
Ankle 0 0 0 0
Fracture 1 0 1 1
Back 0 0 0 0
Right 0 0 0 0
Left 0 0 0 0
Cut 0 0 0 0
Elbow 0 1 1 1
Toe 0 1 1 1
Finger 0 0 0 0
Thumb 0 0 0 0
Eye 0 0 0 0
Ear 0 0 0 0
Throat 1 0 1 1
Leg 0 0 0 0
Hand 0 0 0 0
Lower 0 0 0 0
Trauma 0 0 0 0
Wrist 0 0 0 0
Hip 0 0 0 0
Broken 0 1 1 1
Strain 0 0 0 0
Stress 0 0 0 0
Palm 0 0 0 0
Sore 1 0 1 1
Swollen 1 0 1 1
Ruptured 0 1 1 1
Head 0 0 0 0
Scratch 0 0 0 0
Foot 0 0 0 0
Neck 1 0 ...1 1
Total 6 4 10 10
Distance Measure 3.16 10

i =n
n


The Euclidean distances are calculated using the following formula:
n
d
i,j
= ( (x
i,k
x
j,k
)
2
)
(2)
k=1
where i,j =records and n =number of variables
c) Comparing two medical claims
Record 1 shows sore throat, swollen neck and fractured shoulders whereas Record 2 shows ruptured toe and broken
elbow. The system checks these ailments against the column for variables and the existence of the same is marked with a
one (1) and the absence is marked with a zero (0) as shown in Table 4.5. The absolute difference and corresponding
square are computed for each row. The totals and corresponding Euclidean distances are then calculated. These
distances are the ones that will then be used for clustering.
d) The clustering process and claim analysis process
Medical claim details are entered one by one. When all the working days medical claim data are entered into the system,
the Euclidean distance between each pair of the claim forms is calculated as shown in Figure 4.5. The user of the system
sets the desired maximum distance for the records to belong to the same cluster. The sum claimed for all the claim forms
is computed and the average amount determined. If the amount for a given claim exceeds a set amount for that cluster,
then that particular claim form is rejected. The same process is repeated for different set distances and the rejected claims
is listed.
4.4 Entry of Claim details

Figure 4.5: Entry of Claim details
4.5 Claims Analysis
The Analysis process:
The Analysis process starts when the user clicks the Analyse button shown in Figure 6.3 The program then computes
the Euclidean distance between each pair of the claim forms using the Euclidean distance formula:
n
d
i,j
= |xi,k xj,k| (3)
k=1
where i,j =records, n =number of variables
The user of the system sets the desired maximum distance for the records to belong to the same cluster, Varying the
distance also changes the number of clusters that are formed. The assumption here is that the greater the distance, the
lower the number of clusters for the given data set and vice versa. The sum claimed for all the claim forms is computed
and the average amount determined. If the amount for a given claim exceeds the set amount for that cluster, then that
particular claim form is rejected.
CLAIMS ANALYZER
package com.wakoli.claimsanalyser;
import javax.swing.*;
import javax.swing.event.*;
import java.awt.*;
import java.awt.event.*;
import java.sql.SQLException;
import com.wakoli.claimsanalyser.methods.*;


public class ClaimsAnalyzer extends JFrame{
JPanel
flags =new JPanel();
JLabel
flag[ ]={
new JLabel(new ImageIcon("com/wakoli/claimsanalyser/support/images/flag.gif")),
new JLabel(new ImageIcon("com/wakoli/claimsanalyser/support/images/flag.gif"))},

logo =new JLabel(new ImageIcon("com/wakoli/claimsanalyser/support/images/eck.gif"));
static JDBCAdapter dataModel;
JDesktopPane desktop =new JDesktopPane();Login pass =new Login();
ChangePassword change =new ChangePassword();
boolean changed=false;
boolean loaded=false;
public ClaimsAnalyzer(){
Color background=new Color(150,160,130);
flags.setBackground(background);
flags.setLayout(new GridLayout(8,1));
int i;
for(i=0;i<=7;i++){
//flags.add(flag[i]);
}
ContentPanel contentPane =new ContentPanel("com/wakoli/claimsanalyser/support/images/eck.gif");
desktop.setBackground(background);//(150,160,130));
desktop.add(pass,JLayeredPane.MODAL_LAYER);
getContentPane().add(flags,BorderLayout.WEST);
getContentPane().add(desktop);//Can either add JDesktopPane or set it to be ContentPane
//getContentPane().add(contentPane); //this sets a background image
contentPane.setOpaque(false);
setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
setTitle("THE CLAIMS ANALYZER");
setSize(1024,768);//800,500);
Label c=new Label();setLocationRelativeTo(c);
setUndecorated(true);
setVisible(true);
ChildFrame.select(pass);
//AudioPlayer.play("com/wakoli/claimsanalyser/support/audio/wape vidonge.wav");

Figure 4.6: Claims Analysis


Figure 4.6 shows the dialogue box for clustering and analyzing the claims in which case the distance was set to 10 units
whereby the cluster had 11 records out of which 7 were approved and 4 rejected. Clicking on the View Rejected button
displays a screen as the one shown in Figure 4.7

Figure 4.7: Rejected Claims

Figure 4.7 also shows that the greater the distance, the higher the rejection rate. This is because increasing the Euclidean
distance means having more records within a cluster; hence the chances of netting more fraudulent ones are higher.
5. CONCLUSIONS
This paper shows the successful application of the K-Means clustering algorithm to medical claims records. The medical
claims were successfully clustered and the average amount claimed per cluster was computed. Claims that were far away
from the average were flagged for further scrutiny. Hence the prototype can be used isolate flag suspicious claims that
can be subsequently rechecked. This prototype can immensely increase the medical claim fraud detection rate which in
turn will yield savings that cover operational costs and allowed to increase the quality of the health care coverage, fully
justifying the investment.
REFERENCES
[1]. Anderberg, 1973: Anderberg, M. R. (1973) Cluster Analysis for Applications.
[2]. Andrew Moore: K-Means and Hierarchical Clustering - Tutorial Slides http://www-
2.cs.cmu.edu/~awm/tutorials/kmeans.html (Accessed on 17 December 2008).
[3]. Berry, Michael J. A., and Gordon Linoff (2000), Mastering Data. Mining)
[4]. Brian Everitt, Sabine Landau, Morven Leese, Cluster analysis Edition: 4, illustrated Published by Arnold, 2001 ISBN
0340761199, 9780340761199
[5]. Brian T. Luke: K-Means Clustering http://fconyx.ncifcrf.gov/~lukeb/kmeans.html (Accessed on 23 January 2009).
[6]. DUDA, R. and HART, P. 1973. Pattern Classification and Scene Analysis. John Wiley & Sons, New York, NY.
[7]. Dunn J.C. (1973): "A Fuzzy Relative of the ISODATA Process and Its Use in Detecting Compact Well-Separated
Clusters", Journal of Cybernetics 3: 32-57


[8]. Fayyad, Usama; Gregory Piatetsky-Shapiro, and Padhraic Smyth (1996). "From Data Mining to Knowledge
Discovery in Databases". http://www.kdnuggets.com/gpspubs/aimag-kdd-overview-1996-Fayyad.pdf. Retrieved on
2008-12-17. (Accessed on 23 January 2009).
[9]. Hand D.J., Adams N.M., and Bolton R.J. (eds.) (2002) Pattern detection and discovery. Springer .
[10]. Hans-Joachim Mucha and Hizir Sofyan: Nonhierarchical Clustering
http://www.quantlet.com/mdstat/scripts/xag/html/xaghtmlframe149.html (Accessed on 23 January 2009).
[11]. Hartigan, J. A. (1975). Clustering Algorithms. Wiley. MR0405726. ISBN 0-471-35645-X
[12]. MacQueen, J. (1967). "Some methods for classification and analysis of multivariate observations." In Proceedings
of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, edited by L. M. Le Cam and J.
[13]. Mugenda, O. M. and Mugenda, A. G. (1999) Research Methods: Quantitative and. Qualitative Approaches.
[14]. Paul E. Black, "Euclidean distance", in Dictionary of Algorithms and Data Structures [online], Paul E. Black, ed.,
U.S. National Institute of Standards and Technology. Available from:
http://www.itl.nist.gov/div897/sqg/dads/HTML/euclidndstnc.html (Accessed on 23 January 2009).
[15]. UNESCO, 2004 (Best and Khan, 2004; Mugenda and Mugenda, 1999; Nachmias and Nachmias, 1996).
[16]. Wikipedia, http://en.wikipedia.org/wiki/Euclidean_distance (Accessed on 23 January 2009).
[17]. Yamanishi K., Takeuchi J., G. Williams, and P. Milne, On-line unsupervised outlier detection using finite
mixtures with discounting learning algorithms, In Data Mining and Knowledge Discovery vol. 8, pp. 275300,
2004.

Mr. Leonard .W. Wakoli is a Lecturer, Kenya Methodist University (Department of Computer Science
and Business Information Systems ) Nakuru Campus With well over 10 years of experience. He is also
a Cyber Security Consultant, PhD Candidate (Business Information Systems Security), Jaramogi
Oginga Odinga University Bondo Kenya. He holds a MSc. Software Engineering JKUAT- Kenya,
a Post Graduate Dip. in Management of Information Systems (MIS) Greenwich University UK, a
BSc. in Mathematics and Computer Science JKUAT- Kenya, a Dip. in Science Education KSTC-
Kenya. He is also an Environmental Impact Assessment Auditor , a Motivational Speaker and a Certified
Public Accountant K Finalist. His area of interest in research is Cyber Security, now working on Harnessing the
Power of Intrusion Detection Systems.

Mr. Abkul Orto is a Lecturer in the Department of Information Technology in the School of
Information Technology and Engineering and Director of Open, Distance and eLearning , Meru
University of Science and Technology, Kenya. He holds a Master of Science (Computer Based
Information Systems) from University of Sunderland, United Kingdom. He previously taught at KCA
University for seven years as a full time faculty and also at Institute of Computer Science and
Information Technology and IT Centre of the Jomo Kenyatta University of Agriculture and Technology,
Kenya. His research interests include Human Centred Computing, Spoken Language processing focusing on African
Languages, Information retrieval and security. He has a teaching experience of 12 years research interests include
Human Centred Computing, Spoken Language processing focusing on African Languages, Information retrieval and
security. He has a teaching experience of 12 years.

Mr. Stephen Mageto is a Lecturer in information Technology and Management and Chairman of
Department at Meru University of Science and Technology. He holds a Master of Studies (Information
Technology and Management) degree from Madurai Kamaraj University, 2004. He has over 8 years of
teaching experience. He has a strong research interest in the impact and integration of ICT applications
on community development and strategic information systems.

Application of The K-Means Clustering Algorithm in Medical Claims Fraud / Abuse Detection

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Application of The K-Means Clustering Algorithm in Medical Claims Fraud / Abuse Detection

Transféré par

Droits d'auteur :

Formats disponibles

International Journal of Application or Innovation in Engineering& Management (IJAIEM)

Web Site: www.ijaiem.org Email: editor@ijaiem.org

Vous aimerez peut-être aussi